 Thank you, everyone, for being here. It's a pleasure to present this paper. So the paper I'm presenting is about Learning Dynamics in Tax Punching. And I use administrative tax data from Ecuador. This paper is joint work with Yanemcheguza, a fellow PhD student at the University of Manuel. The motivation of my study is to understand dynamic behavioral responses to tax incentives in a development context. With tax incentives, I mean that tax theory predicts punching at jump points in the marginal tax rate. But there's only limited empirical evidence for actual punch. We look at this in the case of Ecuador, the middle income developing country. And first of all, there's very little evidence of tax punching behavior from developing countries. But most of all, Ecuador is so interesting because it's undergoing the transition from an informal to a formal economy now with a growing number of tax pay. Within this setting, we are able to respond to questions, such as whether people learn about punching or tax adjusting opportunities over time. And more importantly, how this knowledge is transmitted between the two individuals. I don't want to go deep into literature, we draw on two general strands of literature. That's a literature on tax punching, which started with the CIES paper in 2010 and has been used for many purposes. Maybe that's closest to ours is one by Czechian co-authors where they look at knowledge diffusion in US zip code carriers. The second large literature, our paper, our restaurant has won our taxation and development, where I don't want to go much into details. In this paper, we start off by dodging punching behavior in Ecuador. And we analyze learning effects in tax-adjusted opportunities. And we will look at the channels of information transmission, specifically we will look at whether new workers adjust to phone level punching and whether incumbent workers can learn from new co-authors who are punching or not before coming in. As a preview of the results, we see there will be a large spike in taxable income distribution at the first peak. You'll see that this effect is entirely driven by reporting behavior, especially in private production. We will see that punching increases over time and good experience. And we will have a very strong impact of firm level punching rates on the digital side. We will also find some evidence for actual learning going on with the firm level in the digital field. And I will report specifically on this later on. So just in general, some of you may be familiar with this, that there's a large tax punching literature. And the general idea is that there are jumps in the marginal tax rates and these jumps that generate kinks in the budget sets. And standard theory predicts that people haven't incentive to locate at these points of discontinuity. However, I'm here to tell you that this effect is not so pronounced because there are adjustment restrictions and there's lack of knowledge in the sector. One important question in the literature is whether these are reporting effects or real responses like neighbors' response to justice. There's a bit of background on Ecuador. So Ecuador has undergone a number of policy changes in the past years. For our study, most importantly, are some policies to increase tax compliance and formalization. And among these, most importantly for this paper are large-scale deduction possibilities. So individuals have the possibility to deduct expenses of health, education, nutrition, housing, and clothing up to certain limits from their tax liability. We will be focusing on wage earners here. And for wage earners, it's important to keep in mind that the tax declarations are directly submitted by the employer or the given employee. And the deductions, this is a sort of unique setting that the employees even need to report the level of deductions to their employer. They need to report the projected value of deductions at the beginning of the fiscal year. That is then used to calculate the amount of money that is with parents. And only if these deductions exceed the reporting threshold does the employee need to submit an annex with the actual receipts. The data I use, I present here in the study, is from the universe of individual income tax returns from the years 2006 to 2015. These are all firm reported tax forms. We have some limited socio-demographic data of 1% firms, and we will only focus on private sector wage earners. This is the gross income distribution in Ecuador, BIN. So on the horizontal axis, you can see gross income. Here's the number of taxpayers in each BIN. Interestingly, on the right-hand axis, vertical axis, I have the marginal tax rate. You can see that it's about zero up to a gross income of about 10,000 US dollars. This is all real 2013 US dollars. The nominal rate varies every year, but in real terms it stays unchanged. And at $10,000, you start paying taxes and the marginal tax rate jumps from zero to 5%. Then it goes on in smaller steps, actually up to 35%, which is the top marginal tax rate in Ecuador. This is the gross income distribution before using any deductions. The taxable income distribution, so net of any deductions, looks like this. Most notably, you see a large amount of bunching just at this first king, at the exemption threshold. So individuals, they position themselves just below this king so that they do not need to pay any taxes. And this effect is only available at the first king of the tax schedule. So at all further jumps in the tax rate, when it goes from 5% to 10% to 12% to 50% and so on, we find no bunching at this further king. This is the first slide to motivate why we're interested in learning. This is looking at tax avoidance over time. And the blue line is the number of private sector employees with gross income above this first king. So what the exemption pressure is. First of all, you see that there's a strong increase and this is the increase in the number of taxpayers I was talking about. The total number of taxpayers increased by almost, so the total number of tax declarations submitted increased by almost 10 times as much. It goes from about 1 million to 2.5 million. But these are only high wage earning individuals, namely those individuals with gross income above these 10,000 dollars. However, if you look at those people that have taxable income above these 10,000 dollars. So people that actually need to pay taxes, this is the red line, we see that this starting in 2008 when the deduction possibilities were introduced, there was a breach between these two lines. And you see that the gap between them it actually opens up over time. So as time goes on, more people use these tax adjustment opportunities and more people are, yeah, the blue people are those that should, could be paying taxes due to their gross income, the red people are those that are. And we can use the established bunching estimator, which I don't want you to go into detail now, there's a whole literature on that. You can estimate the excess mass, which is about 4.1, which is very large for this literature. That's just saying that in the vicinity of this king point, the mass is about four times as large as it should be due to a counterfactual. We look at these bunching estimates over time. We can see that in 2008 when the reform was introduced, we already have the significant level of bunching. However, these estimates, they increase over time, up to the year 2015 when they're at about six. So over time, these bunching becomes more and more prevalent. If you look at a cohort analysis, and this is what I mean with new people coming into the formal sector, we can look at what happens, say, to individuals who start off in the formal sector in the year 2011. And this whole line shows bunching estimates for this cohort. That means these are individuals that in the year 2011, paid showing up for the first time in the formal taxes. And we see that they bunch a bit in the beginning, but it's not statistically significant in their first year of paying taxes. In 2012, their second year of paying taxes, there's already some bunching going on, not so significant. But these bunching estimators for these cohorts, they increase over time for every cohort. So the longer you have already been in the formal tax system, the more likely you are to bunch. The higher is the estimate for this excess mass. This is the same if we pool people into individuals with no experience, who have not had high income in the past two years. We find moderate amount of bunching. If you look at those individuals with income, high income in the past two years, we find much stronger bunching. And this effect holds true even if we do a probate regression and control for all sorts of socio demographics we have like age, gender, and so on, and education levels. And most importantly also, the actual gross income these individuals earn. I'd like to spend the rest of the presentation to talk about, I've shown you aggregate levels of learning that this information increases over time. Now I would like to talk about how this information is spread. And the first thing we look at this is we look at job switchers and we ask how do job switchers adjust to firm level bunch? And to do so we have an employer employee match data and we can use that to compare workers who move into high bunching versus low bunching environments. And to do so we consider the first switch of the main employer among all job to job transitions in the years 2010 to 2014. That way we can observe at least two years before, consecutive years before moving and after moving. Most importantly, we can assign the old and the new firm to quintiles based on the share of coworkers for a bunch. This is the event study graph I would like to show you and we do lots of regression analysis after this but this is the key result is you can see from this graph. So the key result is that individuals, the green line, individuals moving from a firm in the mid bunching distribution to the high bunching to a firm in the high bunching distribution. This is an event study design. So year zero, we have coded as the first year at the new firm, the red line marks the change. Year minus one is the last year at the old firm. And on the y-axis we have the share of bunchers so the share among these individuals who are actually bunched. And we see that if you move from a firm in the mid quintile to a high quintile, the probability that you bunched strongly increases so that you use these tax-adjusting opportunities. Whereas if you move from a mid to a mid or a mid to a low quintile, the probability that you bunched stays relatively unchanged. This result is what we call learning and memory. You learn from the firm you move into, you learn about these tax-adjusting opportunities. If you move it to a firm where there are a few tax-adjusting opportunities, you remember the opportunities you had and you continue to use them. This results holes through under a number of identification strategies we have. I don't want to go into detail, the general idea is we do an event study design where we're interested in this delta here which is the interaction between an indicator for moving to a high or low quintile and being, having an observation in the year after the move. We control for a number of things. Most importantly, individual fixed effects, time fixed effects, fixed effects in event time and time varying observable characteristics such as wages before and after an individual's move. And the thing is if we go from mid to low, the results say that there's basically no effect. There's a small and modest effect but it goes away once we include controls. Going from mid to high firm, there's a strong effect of three percentage points more likely to bunch and this effect holes through even when controlling for a number of things. We can do the same identification strategy where we now have not just one delta but many deltas and these deltas measure the effect of moving into the high or low bunching firms. And this allows us to estimate anticipatory and post treatment effects. So the results show that moving from a mid to a low bunching environment. In all the years after the person changed the firm to a low firm, there's basically no effect. However, for those individuals moving to a high bunching firm, there's a strong effect and this effect becomes stronger in event year plus one. That means in the second year an individual is working at the new firm. So these effects are persistent and even become stronger in the second year at the new firm which will be in line without thinking if it takes some time to understand and to adapt to firm level bunching behavior and to learn these new methods. Okay, this is basically what I said. So we have a strong and persistent firm level effect asymmetric response and this shows learning and memory as has also been found in other settings in literature. So now we've seen that firms are very important. The natural thing is to ask what determines this firm level bunching? And to do so, I do the same cohort analysis but now I focus on firm cohorts. So I group firms into cohorts by year of the first entry into the formal sector. So by year that the first time the firm has employees who have income above this exemption thresholds. And then I calculate the share of firms within each cohort which have some bunchers, so which have one or more bunchers. And this is the table and basically if you look at this table, again you have cohorts here. So this is the core 2011. These are the share of firms among this cohort who have at least one buncher working in them. And you see that there are basically three things I would like to take. I would like you to take away from this. First of all, a given cohort, the share of firms who have bunchers increases over time as this firm has been in the formal sector. So it goes from 20% here to 31, 38, 41, 53 and so on. And this is throughout all cohorts. Secondly, firms that enter the formal sector later on start so at a higher level. So the firms that started in 2008, they have 20% of individuals who bunch. The firms that started in 2014, they already have 38% of their employees who bunch. Sorry, 38% of these firms who started in the year 2014 in the formal sector have at least one individual who bunches. And lastly, if you look at it column wise, you can see that among each year, firms that have already been in the formal sector for longer are more likely to have bunchers in them than firms that have just entered newly into the formal sector. The question is, do workers learn from new coworkers who were bunching before? And to do so, we compare firms that receive potential bunchers. So individuals who before coming to this firm could have bunched and we put these firms into a treatment group and a control group. Treatment group or individuals who were bunching before, who received individual incoming individuals who were bunching, control group, or firms who received incoming employees who were not bunching before. The short answer is we find basically no effects. Now, it could also be a power issue. There are a few firms that meet these moving criteria we have imposed. It could also be that one individual is not likely to have a strong effect on the incumbent workers. But here it seems that the treatments and the control, there are no significant differences even if we go into small firms where we expect the effects to be strong. So it seems that new coworkers do not have an effect on the bunching share of the incumbent employees. In this paper, I hope I have been able to show you clear evidence for tax bunching behavior in Ecuador driven by reporting behavior by filing detections. I showed that experience with filing taxes increases the probability to bunch. So the more experienced, the longer you've been in the tax system, the more probable you are to use these tax adjustment opportunities. We find a really strong impact of firm level bunching on individual bunching. And this goes on. This is in line with the large literature moment that's currently showing that firms are important for tax avoidance and evasion decisions going on in developing countries. We find evidence for asymmetric adjustments of learning and memory. We also find evidence that the firms themselves are learning so that firms when they've been in the formal sector for longer, they're more likely to have employees who are bunching. But we do not seem to find an effect of new coworkers or incumbent workers at a given firm. Although here we have some power issues in this analysis. Thank you very much.